[FFmpeg-devel] [PATCH V1 1/2] lavfi/vf_scale_vaapi: add scaling mode setting support.

[Li, Zhong]

> -----Original Message-----
> From: ffmpeg-devel [mailto:ffmpeg-devel-bounces at ffmpeg.org] On Behalf
> Of Mark Thompson
> Sent: Monday, December 10, 2018 2:26 AM
> To: ffmpeg-devel at ffmpeg.org
> Subject: Re: [FFmpeg-devel] [PATCH V1 1/2] lavfi/vf_scale_vaapi: add scaling
> mode setting support.
> 
> On 06/12/2018 10:39, Jun Zhao wrote:
> > before this change, scale_vaapi hard coding the scaling mode, add a
> > new option "mode" to setting the scaling mode, it can be use to change
> > scaling algorithm for performance/quality trade off.
> >
> > Signed-off-by: Jun Zhao <mypopydev at gmail.com>
> > ---
> >  libavfilter/vf_scale_vaapi.c |   33
> ++++++++++++++++++++++++++++++---
> >  1 files changed, 30 insertions(+), 3 deletions(-)
> >
> > diff --git a/libavfilter/vf_scale_vaapi.c
> > b/libavfilter/vf_scale_vaapi.c index d6529d5..ad222e6 100644
> > --- a/libavfilter/vf_scale_vaapi.c
> > +++ b/libavfilter/vf_scale_vaapi.c
> > @@ -35,10 +35,26 @@ typedef struct ScaleVAAPIContext {
> >
> >      char *output_format_string;
> >
> > +    int   mode;
> > +
> >      char *w_expr;      // width expression string
> >      char *h_expr;      // height expression string
> >  } ScaleVAAPIContext;
> >
> > +static const char *scale_vaapi_mode_name(int mode) {
> > +    switch (mode) {
> > +#define D(name) case VA_FILTER_SCALING_ ## name: return #name
> > +        D(DEFAULT);
> > +        D(FAST);
> > +        D(HQ);
> > +#undef D
> > +    default:
> > +        return "Invalid";
> > +    }
> > +}
> > +
> > +
> >  static int scale_vaapi_config_output(AVFilterLink *outlink)  {
> >      AVFilterLink *inlink     = outlink->src->inputs[0];
> > @@ -70,6 +86,7 @@ static int scale_vaapi_filter_frame(AVFilterLink
> *inlink, AVFrame *input_frame)
> >      AVFilterContext *avctx   = inlink->dst;
> >      AVFilterLink *outlink    = avctx->outputs[0];
> >      VAAPIVPPContext *vpp_ctx = avctx->priv;
> > +    ScaleVAAPIContext *ctx   = avctx->priv;
> >      AVFrame *output_frame    = NULL;
> >      VASurfaceID input_surface, output_surface;
> >      VAProcPipelineParameterBuffer params; @@ -119,7 +136,7 @@
> static
> > int scale_vaapi_filter_frame(AVFilterLink *inlink, AVFrame *input_frame)
> >      params.output_color_standard = params.surface_color_standard;
> >
> >      params.pipeline_flags = 0;
> > -    params.filter_flags = VA_FILTER_SCALING_HQ;
> > +    params.filter_flags |= ctx->mode;
> 
> "=" would feel safer - "|=" implies something else might have been setting it
> as well.
> 
> >
> >      err = ff_vaapi_vpp_render_picture(avctx, &params,
> output_surface);
> >      if (err < 0)
> > @@ -131,9 +148,10 @@ static int scale_vaapi_filter_frame(AVFilterLink
> > *inlink, AVFrame *input_frame)
> >
> >      av_frame_free(&input_frame);
> >
> > -    av_log(avctx, AV_LOG_DEBUG, "Filter output: %s, %ux%u
> (%"PRId64").\n",
> > +    av_log(avctx, AV_LOG_DEBUG, "Filter output: %s, %ux%u
> > + (%"PRId64"), mode: %s.\n",
> >             av_get_pix_fmt_name(output_frame->format),
> > -           output_frame->width, output_frame->height,
> output_frame->pts);
> > +           output_frame->width, output_frame->height,
> output_frame->pts,
> > +           scale_vaapi_mode_name(ctx->mode));
> >
> >      return ff_filter_frame(outlink, output_frame);
> >
> > @@ -174,6 +192,15 @@ static const AVOption scale_vaapi_options[] = {
> >        OFFSET(h_expr), AV_OPT_TYPE_STRING, {.str = "ih"}, .flags =
> FLAGS },
> >      { "format", "Output video format (software format of hardware
> frames)",
> >        OFFSET(output_format_string), AV_OPT_TYPE_STRING, .flags =
> > FLAGS },
> > +    { "mode", "Scaling mode",
> > +      OFFSET(mode), AV_OPT_TYPE_INT, { .i64 =
> VA_FILTER_SCALING_HQ },
> > +      0, VA_FILTER_SCALING_NL_ANAMORPHIC, FLAGS, "mode" },
> 
> NL_ANAMORPHIC mentioned in this limit but not offered as an option?
> 
> > +    { "default", "Use the default (depend on the driver) scaling
> algorithm",
> > +      0, AV_OPT_TYPE_CONST, { .i64 = VA_FILTER_SCALING_DEFAULT },
> 0, 0, FLAGS, "mode" },
> > +    { "fast", "Use fast scaling algorithm",
> > +      0, AV_OPT_TYPE_CONST, { .i64 = VA_FILTER_SCALING_FAST }, 0, 0,
> FLAGS, "mode" },
> > +    { "hq", "Use high quality scaling algorithm",
> > +      0, AV_OPT_TYPE_CONST, { .i64 = VA_FILTER_SCALING_HQ }, 0, 0,
> > + FLAGS,  "mode" },
> >      { NULL },
> >  };
> >
> >
> 
> LGTM in any case.
> 
> (Ack on keeping the HQ default - IIRC the reason for choosing HQ as the
> mode when this was written was that the default/fast mode was not faster
> and had much worse quality on some of the older Intel platforms (I would
> guess Bay Trail based on what I probably had available at the time).  Might
> be worth investigating further if you have such machines available to test.)

Current implementation is to set VA_FILTER_SCALING_HQ as default (if mode unset), but setting "mode=default" means VA_FILTER_SCALING_DEFAULT.
Thus looks like a conflict and it is confusing.

I think would better to set the default mode to be VA_FILTER_SCALING_DEFAULT, and then driver will choose an appropriate mode. 
E.g: If SVC (Enabled started from Skylake: https://01.org/sites/default/files/documentation/intel-gfx-prm-osrc-skl-vol15-sfc.pdf ) is supported by HW , SVC will be the scaling mode. 
If it is not support, AVS may be choosed.
Implementation in iHD drvier can be referred: https://github.com/intel/media-driver/blob/master/media_driver/linux/common/vp/ddi/media_libva_vp.c#L1138

IMHO It is not a good idea to force scaling with HQ mode no matter what is the implementation of driver and the status of HW.